Small Traffic Sign Detection in Big Images: Searching Needle in a Hay

Yawar Rehman,Hafsa Amanullah,Min Young Kim,Muhammad Ayaz Shirazi

doi:10.1109/access.2022.3150882

Abstract

Traffic sign detection is an essential module of self-driving cars and driver assistance system. The major challenge being, traffic sign appear relatively smaller in road view images. It covers only 1%-2% of the total image area. Hence, its challenging to detect very small traffic sign in a larger image covering huge background of similar shape objects. Thus, we propose YOLOv3 network layers pruning and patch wise training strategy for small sized traffic sign detection. This aids in improving recall percentage and mean Average Precision. We also propose anchor box selection algorithm that uses bounding box dimension density to obtain optimal anchor set for the dataset. This reduces false positives and log-average miss rate. The proposed approach is evaluated on German traffic sign detection benchmark and Swedish traffic sign dataset and proves that it achieved a good balance between mAP and inference time.

Highlights

A MONG several fields on the canvas of artificial intelligence, the intelligent transportation system is the hot research area for the researchers and scientists
The proposed approach is evaluated for six evaluation metrics, namely mean Average Precision, inference time, recall percentage, false positives and log-average miss rate
The network was trained in two stages; for first 10 epochs, Darknet53 network is frozen and rest of the network is trained with the batch size of 32 image patches to obtain a stable loss, and complete network is trained for 50-60 epochs with the batch size of 8 image patches

Summary

Introduction

A MONG several fields on the canvas of artificial intelligence, the intelligent transportation system is the hot research area for the researchers and scientists. The authors in [2] noted that the accuracy of the traffic sign detection system decreases with the increase of distance between the camera and the traffic sign on the road. The authors in [3] noted the same and argued that the detector should be robust to detect the traffic signs of smaller size with reference to the image size. We may conclude that the robust traffic sign detection system should be able to detect the small size traffic signs with reference of the image size This property of the detector will leverage the driver assistance systems in warning the danger ahead of time. The speed of the traffic sign detection system plays an important role These days the convolutional neural network (CNN) provides promising feature set to detect and classify an object. The authors in [7] acquired the upper and lower human body part features from the fully connected layer

Methods

Results

Conclusion